Personal Budget Optimizer

Halfway through junior year, I realized I had genuinely no idea where my money was going. I had a part-time campus job, some freelance tutoring income, and a vague sense that I was “doing fine” financially. Then my checking account hit $47 two days before rent was due.

I tried Mint and YNAB but they felt overengineered for my situation. I wanted something dead simple: import my bank CSV, categorize spending, and tell me where I’m bleeding money. So I built it myself over a weekend with Streamlit.

The core idea is the 50/30/20 rule— 50% of after-tax income to needs, 30% to wants, 20% to savings. Not revolutionary, but actually seeing your numbers mapped against that target is surprisingly motivating.

The app is intentionally minimal. You upload a CSV bank statement (Chase, BofA, and Discover formats supported), and it auto-categorizes transactions using keyword matching on merchant names. Not fancy ML — just a dictionary of patterns like “DOORDASH” → Food Delivery, “SPOTIFY” → Subscription.

From there it calculates your actual 50/30/20 split, flags the biggest deltas between target and actual, and shows month-over-month trends. The whole thing is about 400 lines of Python.

Key features:

• Auto-categorization with manual override for misclassified transactions
• Monthly spending breakdown with Plotly sunburst and bar charts
• Savings rate tracker with goal-setting (I set mine to 20%)
• Subscription detector that flags recurring charges you might have forgotten about

After three months of tracking, the numbers told a pretty clear story.

The biggest surprise was food delivery. I knew I was ordering a lot, but $420/month was genuinely shocking. That’s more than I was spending on groceries. I also found 9 subscriptions I’d completely forgotten about — including a language learning app I signed up for in freshman year and never opened again.

Let me be upfront about what’s hacky here.

• Categorization is keyword-based, not ML. It works for chains (Uber Eats, Target) but struggles with small local businesses or vague merchant names like “SQ *JOHN’S”. About 15% of transactions need manual reclassification.
• No bank API integration. You have to manually download and upload CSV statements each month. I looked into Plaid but the cost didn’t make sense for a personal tool.
• The 50/30/20 rule is a rough heuristic. It doesn’t account for student loan payments, irregular income, or living in a high-cost area. For a student in the Bay Area, 50% on “needs” is pretty tight.
• Single-user, local only. No authentication, no cloud storage. Your data stays in a local SQLite file, which is a privacy feature but also means no access from your phone.

This was my first time building something purely for myself, and it taught me that the best data projects start with a real problem you actually have. I didn’t need a sophisticated model — I needed visibility.

On the technical side, Streamlit is incredible for prototyping. I went from zero to a working dashboard in about 6 hours. The tradeoff is that it’s hard to make it look polished — it always looks like a Streamlit app. But for a personal tool, that’s totally fine.

The biggest personal takeaway: tracking changes behavior. Just the act of seeing my food delivery spending in a bar chart next to my savings was enough to shift my habits. I didn’t need an algorithm to tell me to stop ordering DoorDash three times a week — I just needed to see the number.